Standing Balance of Legged Robots: Leveraging Reduced Order Models to Improve Balancing Performance
Balance , Legged Robotics , Trajectory Optimization , Model Predictive Control , Reduced Order Models
Legged robots have the potential to traverse remote or dangerous areas by utilizing intermittent ground contact to navigate through complex terrain. Despite their promise, legged robots, specifically bipeds, are inherently unstable and fail to operate reliably in unstructured environments. This thesis aims to improve the standing balancing of bipedal robots through the development of a new unified model that incorporates three main fixed base balancing strategies of ankle, hip, and variable height. To evaluate model performance and the role of each strategy, we compared this proposed model to the existing models in push recovery simulations using two optimal control frameworks: trajectory optimization and a nonlinear model predictive control. We also examined the implementation of nonlinear model predictive controllers on a simulated one legged robot. With access to additional balancing strategies, the developed control policies of the unified model showed biological preferences to utilize the available balancing strategies as needed to recover from a large range of disturbances.